LinguaMap: Which Layers of LLMs Speak Your Language and How to Tune Them?

• URL: http://arxiv.org/abs/2601.20009v1
• Date: Tue, 27 Jan 2026 19:38:12 GMT
• Title: LinguaMap: Which Layers of LLMs Speak Your Language and How to Tune Them?
• Authors: J. Ben Tamo, Daniel Carlander-Reuterfelt, Jonathan Rubin, Dezhi Hong, Mingxian Wang, Oleg Poliannikov,
• Abstract summary: We identify and characterize two key failure modes: the multilingual transfer bottleneck and the language consistency bottleneck.<n>We extend logit lens analysis to track language probabilities layer by layer and compute cross-lingual semantic similarity of hidden states.<n>This is the first approach to leverage layer-localization of language control for efficient multilingual adaptation.
• Score: 3.809788214307542
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Despite multilingual pretraining, large language models often struggle with non-English tasks, particularly in language control, the ability to respond in the intended language. We identify and characterize two key failure modes: the multilingual transfer bottleneck (correct language, incorrect task response) and the language consistency bottleneck (correct task response, wrong language). To systematically surface these issues, we design a four-scenario evaluation protocol spanning MMLU, MGSM, and XQuAD benchmarks. To probe these issues with interpretability, we extend logit lens analysis to track language probabilities layer by layer and compute cross-lingual semantic similarity of hidden states. The results reveal a three-phase internal structure: early layers align inputs into a shared semantic space, middle layers perform task reasoning, and late layers drive language-specific generation. Guided by these insights, we introduce selective fine-tuning of only the final layers responsible for language control. On Qwen-3-32B and Bloom-7.1B, this method achieves over 98 percent language consistency across six languages while fine-tuning only 3-5 percent of parameters, without sacrificing task accuracy. Importantly, this result is nearly identical to that of full-scope fine-tuning (for example, above 98 percent language consistency for both methods across all prompt scenarios) but uses a fraction of the computational resources. To the best of our knowledge, this is the first approach to leverage layer-localization of language control for efficient multilingual adaptation.

Related papers

• Layer-Targeted Multilingual Knowledge Erasure in Large Language Models [15.409568435026015]
  We identify intervention depth as the key factor determining multilingual generalization.<n>We propose MUTE, a framework that uses Centered Kernel Alignment (CKA) and Linguistic Regions Development Score (LRDS) to identify intermediate, language-agnostic layers.
  arXiv  Detail & Related papers  (2026-02-26T03:00:07Z)
• Language Steering for Multilingual In-Context Learning [10.932074928744568]
  Large language models' performance on non-English languages remains substantially inferior to English.<n>We propose language vectors -- a training-free language steering approach.<n>We show consistent improvements on multilingual in-context learning over baselines across all tasks and languages tested.
  arXiv  Detail & Related papers  (2026-02-02T16:52:09Z)
• Tracing Multilingual Representations in LLMs with Cross-Layer Transcoders [51.380449540006985]
  Large Language Models (LLMs) can process many languages, yet how they internally represent this diversity remains unclear.<n>Do they form shared multilingual representations with language-specific decoding, and if so, why does performance still favor the dominant training language?<n>We analyze their internal mechanisms using cross-layer transcoders (CLT) and attribution graphs.
  arXiv  Detail & Related papers  (2025-11-13T22:51:06Z)
• Language steering in latent space to mitigate unintended code-switching [1.1330938617817454]
  Large Language Models (LLMs) often exhibit unintended code-switching, reducing reliability in downstream tasks.<n>We propose latent-space language steering, a lightweight inference-time method that identifies language directions via PCA on parallel translations.<n>Our approach mitigates code-switching while preserving semantics with negligible computational overhead.
  arXiv  Detail & Related papers  (2025-10-11T19:49:38Z)
• Lost in Multilinguality: Dissecting Cross-lingual Factual Inconsistency in Transformer Language Models [49.16690802656554]
  We find that Multilingual factual models struggle to provide consistent responses to semantically equivalent prompts in different languages.<n>We propose a linear shortcut method that bypasses computations in the final layers, enhancing both prediction accuracy and cross-lingual consistency.
  arXiv  Detail & Related papers  (2025-04-05T19:43:10Z)
• PolyPrompt: Automating Knowledge Extraction from Multilingual Language Models with Dynamic Prompt Generation [0.0]
  We introduce PolyPrompt, a novel, parameter-efficient framework for enhancing the multilingual capabilities of large language models (LLMs)<n>Our method learns a set of trigger tokens for each language through a gradient-based search, identifying the input query's language and selecting the corresponding trigger tokens which are prepended to the prompt during inference.<n>We perform experiments on two 1 billion parameter models, with evaluations on the global MMLU benchmark across fifteen typologically and resource diverse languages, demonstrating accuracy gains of 3.7%-19.9% compared to naive and translation-pipeline baselines.
  arXiv  Detail & Related papers  (2025-02-27T04:41:22Z)
• CoCo-CoLa: Evaluating and Improving Language Adherence in Multilingual LLMs [1.2057938662974816]
  Large Language Models (LLMs) develop cross-lingual abilities despite being trained on limited parallel data.<n>We introduce CoCo-CoLa, a novel metric to evaluate language adherence in multilingual LLMs.
  arXiv  Detail & Related papers  (2025-02-18T03:03:53Z)
• SLAM: Towards Efficient Multilingual Reasoning via Selective Language Alignment [78.4550589538805]
  We propose an efficient multilingual reasoning alignment approach that precisely identifies and fine-tunes the layers responsible for handling multilingualism.<n> Experimental results show that our method, SLAM, only tunes 6 layers' feed-forward sub-layers including 6.5-8% of all parameters within 7B and 13B LLMs.
  arXiv  Detail & Related papers  (2025-01-07T10:29:43Z)
• Crosslingual Capabilities and Knowledge Barriers in Multilingual Large Language Models [62.91524967852552]
  Large language models (LLMs) are typically multilingual due to pretraining on diverse multilingual corpora.<n>But can these models relate corresponding concepts across languages, i.e., be crosslingual?<n>This study evaluates state-of-the-art LLMs on inherently crosslingual tasks.
  arXiv  Detail & Related papers  (2024-06-23T15:15:17Z)
• How do Large Language Models Handle Multilingualism? [81.15060972112563]
  This study explores how large language models (LLMs) handle multilingualism. LLMs initially understand the query, converting multilingual inputs into English for task-solving. In the intermediate layers, they employ English for thinking and incorporate multilingual knowledge with self-attention and feed-forward structures.
  arXiv  Detail & Related papers  (2024-02-29T02:55:26Z)
• FILTER: An Enhanced Fusion Method for Cross-lingual Language Understanding [85.29270319872597]
  We propose an enhanced fusion method that takes cross-lingual data as input for XLM finetuning. During inference, the model makes predictions based on the text input in the target language and its translation in the source language. To tackle this issue, we propose an additional KL-divergence self-teaching loss for model training, based on auto-generated soft pseudo-labels for translated text in the target language.
  arXiv  Detail & Related papers  (2020-09-10T22:42:15Z)
• Inducing Language-Agnostic Multilingual Representations [61.97381112847459]
  Cross-lingual representations have the potential to make NLP techniques available to the vast majority of languages in the world. We examine three approaches for this: (i) re-aligning the vector spaces of target languages to a pivot source language; (ii) removing language-specific means and variances, which yields better discriminativeness of embeddings as a by-product; and (iii) increasing input similarity across languages by removing morphological contractions and sentence reordering.
  arXiv  Detail & Related papers  (2020-08-20T17:58:56Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.